Electricity prices in real-world markets can vary widely; prices in the New York Independent System Operator’s (NYISO) market, for example, can vary by two orders of magnitude: from a mean of \(\$ \) 30/MWh up to \(\$ \) 4000/MWh. However, simulated price data from electricity system production cost models (PCMs) result in much narrower distributions. This research has developed a new framework to better calibrate simulated price data from PCMs to real-world price data. This framework is based on Bayesian inference and utilizes two different types of Bayesian models to tackle two problems: Bayesian Ridge Regression to capture extreme price spikes and two Dual-head Bayesian Neural Networks (DBNN) to model prices within the normal range. The calibration framework is validated on real-world data from two regional wholesale electricity markets (California Independent System Operator (CAISO) and NYISO). It is shown that the calibrated PCM data much more closely follows the real-world data distributions, achieving closer skewness and kurtosis values and achieving overall improvement in similarity by as much as 73.58%.
